Deep-learning-based beam position estimation with a photon-counting camera in free-space optical communications

The photon-counting camera has single-photon sensitivity and picosecond time resolution, enabling the acquisition of beacons from thousands of kilometers away for deep-space optical sensing and communication. However, the accurate estimation of the beacon beam center is challenged by the random fluctuations of signal photons and the presence of randomly and widely distributed noise photons. In this paper, we propose a deep-learning-based super-resolution beam position estimator (DSRBPE), which improves the accuracy of beam position estimation through reasonable photon data modeling, as well as optimizing the design of the super-resolution convolutional neural network framework and loss function. Simulation and experimental results show that DSRBPE can achieve sub-pixel precision of 0.2 pixels for super-resolution centroid extraction of an extremely weak 4×4 pixel size Gaussian spot on a 32×32 single-photon array detector, and exhibits superior robustness under low signal-to-noise ratio conditions compared with conventional algorithms.